This article has been moved to http://www.technomicon.com/iPad_Saga_-_Week_2.html . Please go to my new web site.
Thanks,
Mark W. Hibben
Sunday, February 7, 2010
Subscribe to:
Post Comments (Atom)
Reviews of the latest developments in consumer electronics and computing.
Mark,
ReplyDeleteWhile your analysis of the PA6T is good, there is another angle that has to be considered. Apple has suggested that almost every iPhone app can run on the iPad, and it takes lot of overhead to convert from the ARM ISA (of the iPhone) to the PowerPC ISA ; Also, the sort of computing power delivered by the PA6T is not really necessary in a portable device like the iPad (portable device chips usually have to consume at most 1 - 2W); From these 2 points, I believe that the A4 is definitely based on the ARM architecture.
Contrary to what is being reported in the media, I strongly feel that Apple A4 has a Cortex-A8 core with a PowerVR SGX 535 inside.
disregarding the compatibility of unmodified existing iPhone apps, there's been clear evidence that PA Semi engineers are working on ARM.
ReplyDeletehttp://www.macrumors.com/2008/09/15/apple-developing-arm-processors-for-iphone/
A former PA Semi employee listed his position at Apple as "ARM CPU architecture team for iPhone".
I, too, side with concerns about the ISA. JIT compiling of all the app store only takes you so far.
ReplyDeleteBut if PA Semi is as crack as we all seem to think, what would stop them from having implemented the ARM instruction set on top of building blocks they were using in the PA6T? They seem to know quite a bit about high performance / low power registers, instruction units, queues, … all the building blocks, as well as gluing them together. They could tune the number of threads, cores, FPUs, DSPs, etc., freely mixing and matching from their licensed and own IP, all tuned to the instruction mix that the iPhone OS, core app library and iTunes store apps throw at it.
My inspiration for all this was my ~1980 Western Digital MicroEngine, which replaced the ISA decoder of the PDP 11-03 with one that implemented the UCSD Pascal assembly language (P-Code). 16 bit bat out of hell.
Just speculating, of course. Fun, eh?